I want to get the instruction mix of my program on Core 2 Duo processor. On Pentiurm 4, I can use VTune to collect the total retired instructions, the floating point instructions, the load, the store, and the branch, then use substraction method to get the integer instructions. Does this methodology make sense?
However, On core 2 duo processor, when using same methodology, I have problem. The sum of memory instruction (load+store) and floating pointer instructions is larger than the total number of retired instructions. So I guess there is some overlap between memory instructions and floating pointer instruction. If so, how I can get accurate instruction mix on Core 2 Duo processor?
Any help is appreciated.
I use the following events to collect floating-point instructions: